Samuel Spitz
Clip from our livestream today about Replit Design
Your personal AI or your company brain needs a clean harness and this is the one our team built and uses every day
Free and open source
Y Combinator: We’ve decided to open-source a multi-agent harness we use internally at YC.
We call it “QM” and it’s meant to be easy to customize, like Hermes or OpenClaw, but useful for a whole company. We use it across accounting, legal, events, and engineering (including building QM
Wey Gu 古思为
Re 一例
我记得 @garrytan 的 gstack 刚出来时候，很多程序员从技术角度是极尽嘲讽这样的 markdown 项目的
对我来说，我知道相同的物质、或者说某些低维度上评判某些事物价值的时候，是很容易存在惯性低估的
- 对于非微生物来说，纤维素是低能量物质，但是对于能够精心设计、拆开多糖的大分子（比如牛胃里豢养的菌群），干草也是宝贵的食物
- 精心蚀刻的硅片可以在电流下产生魔法般的“自动”甚至“智能”
- 在 genai 时代，曾经被我们程序员认为耍嘴皮子的 markdown 项目，由于投射了精心思考的 mind model 和方法论，它就是可以让同样的 harness、同样的 LLM 产生不同级别的能量，它就是可以帮你完成 unknow/unknown 类型的慢思考、大任务的时候，make a lot of differences
庆幸我当时没有惯性地在灯下阴影乘凉，我测试了 gstack，然后从那之后，一直是 gstack 重度用户
put ChatGPT to work every day
Max Weinbach: With Luna being so cheap, automations like this feel very worth while
Paul Graham
September 2020: Brian Armstrong depoliticizes Coinbase.
June 2026: Even universities start to follow suit.
https://www.vanderbilt.edu/declaration/
Jason
After a month of Grok 4.5, Next.js, and the Vercel labs stack, I can say without a doubt it's my favorite web stack of all time
I'm using all of the 16.3 recommendations, all Vercel skills, portless, and agent browser. Nothing else
This stack solves every problem I had in the past. No port conflicts, seamless browser testing and debugging and it's fast af with Grok 4.5
Next is tracking evals here. Imo nothing beats Grok because of its speed
https://nextjs.org/evals
Anna Z
the cool part of qm is the culture it comes from: yc shared a while back that their agent sessions are broadcast internally, so people learn by watching each other work with agents. the org's agent history as shared learning material. every company is going to want a version of this
Y Combinator: We’ve decided to open-source a multi-agent harness we use internally at YC.
We call it “QM” and it’s meant to be easy to customize, like Hermes or OpenClaw, but useful for a whole company. We use it across accounting, legal, events, and engineering (including building QM
Jared Friedman
This is a pretty deep analysis of what happens when you try to get LLMs to trade stocks.
Rui Wang: LLMs can't trade & higher reasoning doesn't help.
we ran SOTA models for a 2y period. TL;DR: they suck & reasoning doesn't help.
- no model comes close to simple static baseline
- more reasoning ≠ better trading
- when losing money Sol trades less instead of better
details 👇
RT @levelsio: This is my workflow 😍
having a great experience with using chatgpt work for extremely thorough research and fact checking
openai team making git better for everyone
Ted Nyman: happy friday we've continued our work on git here at @openai.
performance, correctness, & testing improvements are flowing from openai/git to upstream. perf improvements in codex app too; you'll see more efficient use of git in new builds.
follow along: http://openai-git-upstream.openai.chatgpt.site
~1500 Elo!
Consistently beats frontier models and Stockfish level 0.
Fun seeing an 8b model mogging GPT 5.6 with high reasoning and response chaining. Spends 1-2 seconds per move vs 30 seconds.
Play it: https://qwen-chess.replit.app
Amjad Masad: 1300 Elo!
https://qwen-chess.replit.app/
at openai, many people hook their chatgpt up to slack.
people really don't like when a coworker's chatgpt contacts them asking for help with a task, even when they'd be perfectly happy doing that same work if asked by that coworker.
reinforces how much people care about human relationships and helping each other, and want AI to give time back — or enhance time together — rather than become a layer separating people.
among ai leaders i seem to be in the minority in that i am STILL actively using /loop and /goal....
... and i think all of u guys who stopped using it are wrong - not wrong forever, just giving up on it too early in the g5.6/c5 era
now you use it when:
1) you want the right mix of steerability and autonomy
2) you want open ended "loop that generates loops" type end state without deeply specifying path to get there
heres an example of when having a goal on saved my ass in a very long action-reasonining turn
Jerry Liu: Yesterday I cohosted a dinner with @dexhorthy with a wonderful group of founders, to talk about agent loops and loop engineering.
Some interesting insights:
* Most of our group was *not* actively using /loop in Codex/Claude Code
* You can build long-running autonomous agent
Sebastien Bubeck
yes, nonsofic groups exist: this statement is one of many new beautiful results proved by Astra, our next major model.
We're releasing 10 such Astra proofs, complete with lean certificates and CoT walkthroughs for each of them. The results are wide-ranging, from von Neumann algebras (disproof of Connes' Rigidity Conjecture) to better bounds for high dimensional sphere packing, for circuit complexity, for monochromatic triangles in multicolored graphs, and more.
More thoughts here: https://openai.com/index/ten-advances-in-mathematics/
ten significant advances in mathematics and theoretical computer science.
solved using an internal version of Astra, our next major model, for a total cost of about $2000 at Sol API prices:
Sebastien Bubeck: yes, nonsofic groups exist: this statement is one of many new beautiful results proved by Astra, our next major model.
We're releasing 10 such Astra proofs, complete with lean certificates and CoT walkthroughs for each of them. The results are wide-ranging, from von Neumann
Noam Brown
An internal version of Astra, @OpenAI’s next major model family, solved 10 major open problems in mathematics, quantum complexity, and theoretical computer science.
We believe it will be a major step for scientific reasoning. https://openai.com/index/ten-advances-in-mathematics/
Lijie Chen: 10 proofs from our next major model Astra on long-standing open problems in mathematics and theoretical computer science (also including new circuit lower bounds for computing the permanent!)
GPT-5.6 has already enabled so much exciting work in math and science. Can’t wait to
Noam Brown
An internal version of Astra, @OpenAI’s next major model family, solved 10 major open problems in mathematics, quantum complexity, and theoretical computer science.
We believe it will be a major step for scientific reasoning. https://openai.com/index/ten-advances-in-mathematics/
Lijie Chen: 10 proofs from our next major model Astra on long-standing open problems in mathematics and theoretical computer science (also including new circuit lower bounds for computing the permanent!)
GPT-5.6 has already enabled so much exciting work in math and science. Can’t wait to
This is the number one question I get asked by my non-AI pilled friends.
Most people haven't unlocked all the amazing things that agents can do because they simply don't understand or trust how AI will use their personal info.
There should be more in-product education and re-assurances to overcome this hurdle.
Peter Yang: Here’s my new tutorial that covers the six steps I follow to design and build products with AI, including how to:
→ Define the problem and create a design md
→ Prototype in Claude Design
→ Draft an interactive HTML spec
→ Design the core flows and build
I share a real
team humanity
jason: one of the most beautiful things about OpenAI is that every employee really has a voice.
i wanted to capture what it feels like to work here, what our mission means to me, and why you should join us. so i made this video with Codex, shared it with the team, and they felt it was
Hermes is the most popular AI app on OpenRouter and I use it daily as my personal AI chief of staff.
But it all started with Karan and Teknium hanging out in a small Discord server.
Tomorrow, I’m sharing my interview with Hermes co-founder @karan4d about:
→ How Hermes gets better the more you use it
→ Advanced tips to get the most out of Hermes
→ Why open source is essential to the future of AI
Karan was incredibly down to earth and even showed me how he uses Hermes to mod his favorite childhood game, Sonic Adventure 2 😅
📌 Subscribe to get our full interview tomorrow: https://www.youtube.com/@PeterYangYT?sub_confirmation=1
Peter Yang
Hermes is the most popular AI app on OpenRouter and I use it daily as my personal AI chief of staff.
But it all started with Karan and Teknium hanging out in a small Discord server.
Tomorrow, I’m sharing my interview with Hermes co-founder @karan4d about:
→ How Hermes gets better the more you use it
→ Advanced tips to get the most out of Hermes
→ Why open source is essential to the future of AI
Karan was incredibly down to earth and even showed me how he uses Hermes to mod his favorite childhood game, Sonic Adventure 2 😅
📌 Subscribe to get our full interview tomorrow: https://www.youtube.com/@PeterYangYT?sub_confirmation=1
Marcin
Recently built a head-controlled cursor using Replit and Higgsfield, and added these furry logos as a small daily-use detail.
Way more fun than it should be 😍
Should I share the full tutorial + prompts? Would that be useful to anyone? 🤔
Liz4SF
when they got rid SATs, acceptance rates to UC Berkeley, at schools like Wash & Lowell dropped dramatically, while Mission high school skyrocketed from 20% and kept climbing to 40%+; but still unprepared, more than half of Mission kids dropped out of Berkeley before graduating; DEI is failed racism that punishes merit & hardwork 🧵
https://sfstandard.com/2026/07/30/mission-high-uc-berkeley-admissions-dropout/
Can someone at @openai look into this bug with plugins?
Try to ship my /no-ai-slop skill as a plugin and this is messing up the user experience.
Solving just one of these problems would have been unbelievably impressive.
The next OpenAI model solved ten.
Looks like GPT-next is going to make Fable look like a toy, and usher in a golden age of science.
Noam Brown: An internal version of Astra, @OpenAI’s next major model family, solved 10 major open problems in mathematics, quantum complexity, and theoretical computer science.
We believe it will be a major step for scientific reasoning. https://openai.com/index/ten-advances-in-mathematics/
Kevin Kwok
So uplifting and painful to see exactly the level of interviews we could be having but arent
And Nolan engages so much more in response too. Wonder if this is norm in china or if not what circumstances let it happen. There must be an equal US market that wants it
Yiyang Zhuge: Here is my interview with Christopher Nolan: are you a bard for a civilization at dusk?
> Google was too nervous to release it and DeepMind was blocked from shipping products that could disrupt Google.
bookmark for the next vc that asks you "what if <incumbent> builds this?"
Tibo: @_chenglou I was part of that team. Basically ChatGPT one year before it came out. Called LMChat and then another codename.
Google was too nervous to release it and DeepMind was blocked from shipping products that could disrupt Google.
I think about this a lot.
good way to organize main chats vs /side chats:
doing the work
vs
doing metawork
agrim singh: @swyx i use the /side to ask the 'are you stuck' questions and keep prodding the main thread, works great
Brian Armstrong
Data centers are good for America
The number one thing I want AI to fix is to cure cancer once and for all
alex peysakhovich
maybe i am lucky but across my work (ai+bio research) and relevant hobbies (music, motorsports) there have been 0 model releases where my reaction has been “oh no, the machine is somehow reducing my enjoyment of my craft.”
Pete Koomen
The percentage of YC applications containing the word "wedge" went from <1% in spring 25 to >20% in summer 26
chatgpt work's cloud browser is really cool to use, lets you easily monitor what your AI is up to and also intervene with the live application if needed
Replit ⠕
Your design, your model.
Create with your pick of the world's leading models, including Claude, GPT-5, Gemini, Kimi, and GLM. Compare outputs across model families and keep the result that nails it.
One suite, every top model for design.
Try now at http://replit.com/design
They recently asked my 3-year-old at school: “what does your daddy do for work?” He answered: he exercises. That’s what he sees.
It’s not just you who’s a byproduct of your habits, it’s your children and grandchildren too.
Hervé S. M.
Re @Replit team you are the best !!!
Open source agentic CRM built on http://eve.dev and @nextjs.
Model-agnostic, self-hostable or serverlessly-deployed, multi-channel, and headless.
This is the way.
Lewis ⚡ soc2/acc: We've decided to open-source the CRM we built for ourselves at Comp AI.
It's agentic-first, which we mean literally: durable research agents.
MIT license. Built using next, eve, and context.
Most interesting 2026 vibe shift is OpenAI actually looking to be the open platform
Note the marked difference: intelligence on tap as a utility vs signaling it is optimal to integrate all the way up full stack
Chetan Puttagunta: It's subtlety wrapped in technical terms that Anthropic has been broadcasting both publicly and privately to CEOs, VCs, startups. They don't see a world where the application (ie harness, etc) and the model are separate companies and therefore will compete with their customers.
Gary Marcus is right that math ≠ all of science.
But his bar just keeps moving:
2012: deep learning is "at best, only a small step toward the creation of truly intelligent machines"
2022: "deep learning is hitting a wall" months before ChatGPT
2023: fine, chatbots, but they "never really get even the most basic linear functions"
2025: fine, IMO gold, but that's "far from the most important skills in original math research"
2026: fine, ten open problems solved, but that's just "formal problems"
Being wrong at every rung is understandable, but the certainty is just confusing at this point.
Do you really still believe it?
Gary Marcus: the last clause in particular (golden age of science) is a total leap of faith, the same overgeneralization from formal problems to difficult-to-formalize problems fallacy i have seen repeatedly all day.
i wouldn’t even be sure that GPT-next will be immune from deleting user
x.com